groupLearning.jl

groupLearning.jl is a Julia package for performing many-to-many and many-to-one transfer learning on feature sets of multiple domains.

Description

This repository implements the Group Alignment Algorithm (GALIA) method [1]. All necessary code and some example data are provided in order to test and replicate study [1].

Getting Started

Dependencies

• Make sure the code is running on Julia release v1.9.0.
• All the required packages (with compatible versions) are provided in Project.toml file.

Installing

• Download the repository.
• Run Julia REPL in the top folder of the repository.
• Instantiate the environment by running

]instantiate

Notes

• Machine learning models are imported from the scikit-learn Python library. Therefore, Python must be installed on the PC. We suggest to follow the steps on ScikitLearn.jl page.

• As an example, recordings of 8 subjects (three session from each) are provided under the folder exampleData/bi2015a.

• Other databases can be tested as long as they are in the NY file format. Please locate custom databases as subfolders under data folder. For other formats you will have to replace the relevant EEG reading functions with your own.

• Please make sure the current working directory is set to the filepath of the repository.

Running the Group Learning

• Open the test_pipeline.jl file.

• Create required objects for pipelines

# Subject specific train/test pipeline
obj_list = initiateObjects(dbName, filepath);

• Create a list comprised of pipeline steps.

# Group learning pipeline
pipeline2 = [createTSVectors, prepareGL, runGL, trainGL];

• Run the pipeline

runPipe!(pipeline2, obj_list)

• Plot and compare pipelines

# PLot and compare pipelines
plotAcc(obj_list)

Authors

Fatih Altindis is a Research Assistant at Abdullah Gul Univeristy, Kayseri. contact: fthaltindis at gmail dot com

Marco Congedo, is a Research Director of CNRS (Centre National de la Recherche Scientifique), working at UGA (University of Grenoble Alpes).

The research on GALIA has has been carried out during a visit of Fatih in Grenoble.

License

This project is licensed under the MIT License

References

[1] Altindis F., Banerjee A., Phlypo R., Yilmaz B., Congedo M. (2023) Transfer Learning for Brain-Computer Interfaces by Joint Alignment of Feature Vectors,IEEE Journal of Biomedical and Health Informatics.

[2] Congedo M., Bleuzé A., Mattout J. (2022) Group Learning by Joint Alignment in the Riemannian Tangent Space GRETSI conference, 6-9 September 2022, Nancy, France.

[3] Congedo M., Phlypo R., Chatel-Goldman J. Orthogonal and non-orthogonal joint blind source separation in the least-squares sense. (2012) The 20th European Signal Processing Conference (EUSIPCO), 27-31 August 2012, Bucharest, Romania.